David decided: Per-document podcast via direct TTS, not the full podcast pipeline

David specified a new UI element inside workspace documents and pushed back on any plan that would route through the existing podcast generation flow:

go study how the lecture ntoes workspace podcast generation system works. i want to build the following: inside of a workspace document, near the top next to the test memo objectives buttons, i want a UI element button for "listen to this lecture" where, if clicked, it generates the podcast just for that document. except, it shouldnt go through the typical kerra podcast generation flow.

He also called out failure modes the agent's initial plan hadn't accounted for:

okay wait but your plan doesnt seem to account for race conditions, failure modes, recoveries, retrying, not having UI states become stale and nto actually indicative of truth etc